Robot Hand Learns How To Learn From Babies

jcasman writes "Wired's got a piece on building a better robotic hand at Stanford. The new robot is called Stair 1.0, and scientists are hoping to take a cue from human children for how to teach a robot to learn. 'When a computer fails at a task, it spouts an error message. Babies, on the other hand, just try again a different way, exploring the world by grabbing new objects -- shoving them into their mouths if possible -- to acquire additional data. This built-in drive to explore teaches us how to use our brains and bodies. Now a number of hand-focused roboticists are building machines with the same childlike motivation to explore, fail, and learn through their hands.'"

Comments

Robot grabbing and shoving things in its mouth

[JetlagMk2]

Am I the only one worried here?

Re:Robot grabbing and shoving things in its mouth

[ParaShoot]

"Hello human, what are you?"

*nom nom nom*

About time!

[sm62704]

'When a computer fails at a task, it spouts an error message. Babies, on the other hand, just try again a different way, exploring the world by grabbing new objects -- shoving them into their mouths if possible -- to acquire additional data

Access gives me the most amusing error messages. "Error 3417: there is no message for this error" (the message is real, the number I pulled out of my ass).

But thinking about it, a robot looking for better data might be a good idea, but a computer? That might worrry me.

Don't forget that a computer, even one running a robot, is just an alectronic abacus, nothing like a human or any other animal's brain. The temptation is to anthropomorphise.

-mcgrew

Re:About time!

[spun]

But a brain is more than likely, and to the best of our current scientific understanding, just a really, really complicated abacus.

Related

[UbuntuDupe]

This reminded me of this video that I favorited on youtube, in which a robot is "brought to life" and then "feels around" to model the world and itself, and then "figures out" how to walk.

This seems really interesting and something I'd want to work on. Anyone know what I would need to learn and do in order to get involved on a theoretical or practical level?

Re:Related

[Stefanwulf]

Anyone know what I would need to learn and do in order to get involved on a theoretical or practical level?

The first thing I'd do is get a feel for the field. MIT's OpenCourseWare has some interesting stuff like Intro to Robotics. I've actually also found wikipedia to be helpful in determining the areas of specialization within a field, and some of the basic jargon that has developed. There will be certain levels of math, programming, and possibly physics or electrical engineering knowledge that you'll need to have, and I've found one of the best ways to get a feel for those is to go to a local university library and pull some papers/conference proceedings on the subject. Read a few of them (from different people, and preferably different conferences/journals) that have titles which interest you, and take note of things like the level of mathematics or engineering knowledge being applied. If you don't understand it, don't be discouraged...it's just a technical language used in papers, and it's not terribly hard to learn, especially when you're learning it within a directly applied framework. It's worth it too, being able to quickly and easily read the papers being published lets you benefit from a huge realm of work that other people are already doing.

On a hobbyist level of involvement, my impression is that's it's far cheaper to get involved with the AI/control side of robotics than it is the hardware design. There are a number of freely/cheeply available robot simulators - some are listed at http://www.robotcafe.com/dir/Software/Simulators/ and http://www.google.com/Top/Computers/Robotics/Software/Simulation/. These let you play with the control systems without having to worry about constructing/purchasing the hardware. Alternatively, for less than $1000 US, you can set up a fairly cheap robotics lab with an AIBO, a wireless connection to your computer, and software like Tekkotsu or URBI.

And of course, if some aspect of it really catches your interest and you want to pursue it professionally, your best bet is probably to start looking at studying with the academic departments which have been publishing the papers/materials you've enjoyed the most, or which have strong programs in that area of the field.

Re:Not too useful

[Nos.]

Simple, you learn how they learn. Children learn alot in the first few years of their life, and the rate at which they learn can be quite astonishing.

Re:With the whole...

[Frequency Domain]

...I sure hope they're not testing it out on babies.

No, they use lawyers and politicians. They were using rats, but too many people protested.

RobotCub - open source hardware

[rich_from_shadow]

RobotCub is particularly interesting because all the design is being posted on their website, so anyone with a spare 200kEuro can build one. It's an EU-funded project, and it's good to see government money widening the pool of Open Source stuff - see www.robotcub.org

Old News

[Agilus]

Seven years ago, when I started grad school at UMass Amherst, they were talking about this stuff in the robotics lab. This is hardly new stuff.

One of the ideas I got out of their talk about their research was that babies start using their muscles gradually, with only a few degrees of motion to start, and moving up from there. For instance, you'll notice that when a baby starts using his hands, he just grabs things for a while. There is no fine grained control of the fingers, and even the control necessary to consciously release whatever he grabbed doesn't come until later.

Building a better?

[rich_from_shadow]

Not quite sure why it's a better robot Hand - Barrett have been around for some time, and their three-fingered system is good, but it's an interesting gripper. The people in the article are the customers, not the hand developers. Still, interesting research is interesting research ObDisclaimer - Shadow does five-fingered hands...

Re:Not too useful

[moderatorrater]

Too bad it doesn't work on Britney Spears...

Let me be the umpteenth...

[Nerdposeur]

...to welcome our drooling, pooping robot overlords.

Re:baby programmers?

[Oktober Sunset]

well at least they have an excuse for living with thier parents...